Expose TORCH_XPU_API to Pytorch

Signed-off-by: xinan.lin <[email protected]>
intel · May 21, 2024 · a1a2448 · a1a2448
1 parent 5bf9e0c
commit a1a2448
Show file tree

Hide file tree

Showing 3 changed files with 17 additions and 6 deletions.
diff --git a/src/aten/CMakeLists.txt b/src/aten/CMakeLists.txt
@@ -1,10 +1,19 @@
 # ATen XPU sources
 
 file(GLOB xpu_cpp "*.cpp")
+file(GLOB xpu_h "*.h")
 file(GLOB xpu_sycl "sycl/*.cpp")
 
 list(APPEND ATen_XPU_CPP_SRCS ${xpu_cpp})
 list(APPEND ATen_XPU_SYCL_SRCS ${xpu_sycl})
 
 set(ATen_XPU_CPP_SRCS ${ATen_XPU_CPP_SRCS} PARENT_SCOPE)
 set(ATen_XPU_SYCL_SRCS ${ATen_XPU_SYCL_SRCS} PARENT_SCOPE)
+
+set(INCLUDE_DIR_FOR_TORCH "${CMAKE_BINARY_DIR}/aten/src/ATen/xpu" )
+file(MAKE_DIRECTORY ${INCLUDE_DIR_FOR_TORCH})
+foreach(HEADER  ${xpu_h})
+  file(COPY ${HEADER} DESTINATION ${INCLUDE_DIR_FOR_TORCH})
+  # uncomment the following line to install the header to torch/include/ATen/xpu
+  # install(FILES ${HEADER} DESTINATION "include/ATen/xpu")
+endforeach()
diff --git a/src/aten/EmptyTensor.cpp b/src/aten/EmptyTensor.cpp
@@ -1,6 +1,7 @@
 #define TORCH_ASSERT_NO_OPERATORS
 #include <ATen/Context.h>
 #include <ATen/EmptyTensor.h>
+#include <ATen/xpu/EmptyTensor.h>
 #include <c10/core/DeviceGuard.h>
 #include <c10/xpu/XPUCachingAllocator.h>
 

diff --git a/src/aten/EmptyTensor.h b/src/aten/EmptyTensor.h
@@ -4,37 +4,38 @@
 namespace at::detail {
 
 // XXX: add TORCH_XPU_API
-TensorBase empty_xpu(
+TORCH_XPU_API TensorBase empty_xpu(
     IntArrayRef size,
     ScalarType dtype,
     c10::optional<Device> device_opt,
     c10::optional<c10::MemoryFormat> memory_format_opt);
 
-TensorBase empty_xpu(
+TORCH_XPU_API TensorBase empty_xpu(
     IntArrayRef size,
     c10::optional<ScalarType> dtype_opt,
     c10::optional<Layout> layout_opt,
     c10::optional<Device> device_opt,
     c10::optional<bool> pin_memory_opt,
     c10::optional<c10::MemoryFormat> memory_format_opt);
 
-TensorBase empty_xpu(IntArrayRef size, const TensorOptions& options);
+TORCH_XPU_API TensorBase
+empty_xpu(IntArrayRef size, const TensorOptions& options);
 
-TensorBase empty_strided_xpu(
+TORCH_XPU_API TensorBase empty_strided_xpu(
     IntArrayRef size,
     IntArrayRef stride,
     ScalarType dtype,
     c10::optional<Device> device_opt);
 
-TensorBase empty_strided_xpu(
+TORCH_XPU_API TensorBase empty_strided_xpu(
     IntArrayRef size,
     IntArrayRef stride,
     c10::optional<ScalarType> dtype_opt,
     c10::optional<Layout> layout_opt,
     c10::optional<Device> device_opt,
     c10::optional<bool> pin_memory_opt);
 
-TensorBase empty_strided_xpu(
+TORCH_XPU_API TensorBase empty_strided_xpu(
     IntArrayRef size,
     IntArrayRef stride,
     const TensorOptions& options);